The Agent Skills Directory

[SAFE]: The skill demonstrates standard multi-agent coordination patterns, including explicit persona framing and behavior constraints to limit agent actions to research and reporting.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to audit external data, which introduces a potential attack surface for indirect prompt injection if the project being reviewed contains adversarial content.
Ingestion points: Subagents ingest source code from a local [PATH] and site content via a [URL] using browser tools (SKILL.md).
Boundary markers: The prompt templates do not currently employ specialized delimiters or boundary markers to isolate analyzed code from the review instructions.
Capability inventory: The skill restricts subagent actions to analysis only. The prompt template explicitly commands: 'Do NOT write any code — only research and report findings.'
Sanitization: The skill relies on the core LLM's instruction-following capabilities and the 'no-code' directive rather than programmatic content sanitization.
[COMMAND_EXECUTION]: The skill uses a parallel dispatch pattern via the Agent tool with the run_in_background parameter. This is a platform-standard method for agent orchestration and does not represent arbitrary or dangerous shell command execution.

experts